# Port to serve the application on the host (routed via Nginx) APP_PORT=8000 # GPU index to allocate (e.g. 0, 1, or 0,1) CUDA_VISIBLE_DEVICES=0 # Run layout parser pipeline on GPU since vLLM VRAM allocation has been tuned/reduced PIPELINE_DEVICE=gpu:0 # Secret used to sign/verify account login JWTs (dev-only value; replace for any real deployment) JWT_SECRET=dev-only-insecure-secret-change-me